Machine learning techniques for structural health ?· Machine learning techniques for structural health…

Download Machine learning techniques for structural health ?· Machine learning techniques for structural health…

Post on 28-Jun-2018

212 views

Category:

Documents

0 download

Embed Size (px)

TRANSCRIPT

<ul><li><p>Machine learning techniques for structural health monitoring </p><p>Kay SMARSLY, Kosmas DRAGOS and Jens WIGGENBROCK </p><p>Chair of Computing in Civil Engineering, Bauhaus University Weimar, Coudraystr. 7, 99423 Weimar (Germany) kay.smarsly@uni-weimar.de </p><p>Key words: Structural health monitoring, machine learning, sensor fault detection, analytical </p><p>redundancy, computer-aided structural assessment </p><p>Abstract </p><p>Data-driven approaches are particularly useful for computer-supported assessment of civil </p><p>engineering structures (i) if large quantities of sensor data are available, (ii) if the physical </p><p>characteristics of the structure are complex to model (or even unknown), or (iii) if the </p><p>computational efforts are to be reduced. This paper, upon a classificational review of </p><p>machine learning techniques in structural health monitoring, reports on an embedded </p><p>machine learning approach for decentralized, autonomous sensor fault detection in wireless </p><p>sensor networks, facilitating reliable and accurate structural health monitoring. Based on </p><p>decentralized artificial neural networks, the embedded machine learning approach is applied </p><p>to perform autonomous detection of sensor faults injected in the acceleration response data </p><p>collected by a prototype structural health monitoring system. As demonstrated through </p><p>laboratory tests, the results highlight the ability of the embedded machine learning approach </p><p>to autonomously detect sensor faults in a decentralized manner, thus enhancing the </p><p>reliability and accuracy of structural health monitoring systems. </p><p>1 INTRODUCTION </p><p>Advancements in sensor technologies have enabled economically affordable sensor </p><p>installations for long-term monitoring of civil engineering structures. Structural health </p><p>monitoring involves installations of hundreds to thousands of sensors to collect valuable data </p><p>about the structure. With increasing complexity and heterogeneity of sensor data, data </p><p>integration and data analysis have become important issues for decision making with respect </p><p>to diagnosis of the structural condition and the prognosis of structural damage [1, 2]. </p><p>Data analysis in structural health monitoring, from a computer science perspective, aims at </p><p>transforming sensor data into useful information and probably into knowledge about the </p><p>structure. The information and knowledge gained from the sensor data is then used for </p><p>structural assessment and for decision making in several respects, such as life-cycle </p><p>management [3] or lifetime prediction [4]. Two general approaches exist for assessing the </p><p>structural condition of civil engineering structures, physics-based approaches and data-driven </p><p>approaches [5]. Physics-based approaches establish first-principle models, mapping the </p><p>physical characteristics of the structure (e.g. using finite element analysis), and then compare </p><p>the outputs of the physical models with sensor data obtained from the monitored structure in </p><p>order to assess the structural condition [6]. Although significant efforts have been undertaken </p><p>to render physics-based models more efficient in terms of computational performance, for </p><p>example for embedment into resource-constraint wireless sensor nodes [7, 8], physics-based </p><p>8th European Workshop On Structural Health Monitoring (EWSHM 2016), 5-8 July 2016, Spain, Bilbao</p><p>www.ndt.net/app.EWSHM2016M</p><p>ore </p><p>info</p><p> abo</p><p>ut th</p><p>is a</p><p>rtic</p><p>le: h</p><p>ttp://</p><p>ww</p><p>w.n</p><p>dt.n</p><p>et/?</p><p>id=</p><p>1982</p><p>8</p></li><li><p>2 </p><p>approaches are generally more computationally intensive than data-driven approaches. </p><p>Data-driven approaches also establish models for comparison with sensor data, but data-</p><p>driven models exploit information from previously collected sensor data, referred to as </p><p>training data [9]. While physics-based approaches are valid in a large operating range </p><p>without the need for extensive quantities of sensor data, data-driven approaches allow </p><p>learning patterns in the sensor data without any knowledge on the physical characteristics of </p><p>the structure [10]. Data-driven approaches are particularly useful, if (i) large quantities of </p><p>sensor data are available, (ii) the physical characteristics of the structure are complex to </p><p>model (or even unknown), or (iii) the computational efforts are to be reduced. </p><p>A variety of data-driven approaches, particularly machine learning techniques, has been </p><p>proposed in structural health monitoring (SHM) for assessing civil engineering structures. </p><p>Machine learning in the context of SHM can be described as the task of generating </p><p>knowledge from past experiences (or, more precisely, from collected sensor data), focusing </p><p>on the prediction of new sensor data. While in artificial intelligence research machine </p><p>learning techniques have been studied since many decades (e.g. for robot control, human-</p><p>computer interaction, or speech recognition), its importance in SHM applications </p><p>substantially continues to grow since about 20 years [11, 12]. For example, Worden and </p><p>Manson [13] have illuminated the utility of machine learning to damage identification, </p><p>concluding that neural networks are still popular, and systems like support vector machines </p><p>are beginning to appear more regularly. Figueiredo et al. [14] have investigated auto-</p><p>associative neural networks, factor analysis, Mahalanobis distance, and singular value </p><p>decomposition to study operational and environmental variability and its influence on </p><p>damage detection of civil engineering structures. Dervilis [15], centered on SHM of wind </p><p>turbine blades, also explores auto-associative neural networks and formulates pattern </p><p>recognition algorithms. In addition, robust multivariate statistical methods are introduced to </p><p>account for the influence of operational and environmental variation on damage-sensitive </p><p>features; the algorithms described are the Minimum Covariance Determinant Estimator and </p><p>the Minimum Volume Enclosing Ellipsoid. Park et al. [16], also focusing on wind energy </p><p>research, couple Gaussian Discriminative Analysis and Gaussian Mixture Models to analyze </p><p>and to predict wind turbine loads in various atmospheric conditions. Nick et al. [17], </p><p>reporting significant trade-offs between accuracy and runtime of the machine learning </p><p>techniques proposed, have used unsupervised learning for identifying the existence and </p><p>location of damage (k-means and self-organizing maps) and supervised learning for </p><p>identifying the type and severity of damage (support vector machines, naive Bayes </p><p>classifiers, and feed-forward neural networks). </p><p>This paper presents an embedded machine learning approach for decentralized, </p><p>autonomous fault detection in wireless SHM systems. Sensor faults and miscalibrations </p><p>substantially affect sensor data and may compromise the reliability and accuracy of SHM </p><p>systems. Specifically in data-driven approaches, the integrity of the sensor data needs to be </p><p>preserved to enhance the reliability and accuracy of SHM system outputs as well as the </p><p>robustness of algorithms implemented for structural health monitoring. In the study reported </p><p>in this paper, the efficient detection of sensor faults and miscalibrations is based on the </p><p>correlations among the response data of different sensor nodes, referred to as analytical </p><p>redundancy, which is implemented through an embedded machine learning approach based </p><p>on artificial neural networks. This paper is organized as follows: First, an overview of </p><p>machine learning techniques commonly used in structural health monitoring is provided. </p><p>Then, the embedded machine learning approach for decentralized, autonomous sensor fault </p><p>detection, based on artificial neural networks, is implemented into a wireless SHM system. </p></li><li><p>3 </p><p>Serving as a testbed for the proposed approach, a laboratory test structure is used in this paper </p><p>for validation, followed by a concise summary of the study presented herein. </p><p>2 AN EMBEDDED MACHINE LEARNING APPROACH FOR DECENTRALIZED, </p><p>AUTONOMOUS SENSOR FAULT DETECTION </p><p>In computer science and in computational engineering, the process of detecting patterns </p><p>and structures within data sets is commonly known as data mining. The detection of patterns </p><p>enables future predictions and decision making, while representing the patterns in terms of </p><p>structures facilitates the extraction of conclusions on the patterns. In data mining, the </p><p>techniques employed to detect patterns within data sets fall into the category of machine </p><p>learning. </p><p>As mentioned previously, due to the computational burden of physics-based approaches in </p><p>structural health monitoring, data-driven approaches, such as machine learning, have been </p><p>gaining increasing attention. In SHM, machine learning is understood as the task of </p><p>generating knowledge about the structural behavior from previously collected sensor data. </p><p>While structural responses are theoretically well explained and documented, the detection of </p><p>such responses in full-scale structures is non-trivial due to the complex nature of actions and </p><p>the actually unknown properties of the structure. Furthermore, SHM outputs may be affected </p><p>by sensor faults and miscalibrations, which may be hardly visible in the collected data. In this </p><p>context, machine learning is applied to detect such hidden, non-evident, or inadequately </p><p>described phenomena. In this section, the machine learning techniques typically applied in </p><p>SHM are briefly discussed. Then, an embedded machine learning approach for decentralized, </p><p>autonomous detection of sensor faults and miscalibrations is presented. </p><p>2.1 Classification of machine learning techniques for structural health monitoring </p><p>Machine learning techniques can be classified into three broad categories according to the </p><p>nature of learning: 1) supervised learning, 2) unsupervised learning, and 3) semi-supervised </p><p>learning [18]. Supervised learning provides a learning scheme with labeled data, i.e. </p><p>examples that include specified outputs (pairs of input data and output data). Using labeled </p><p>data, rules are developed in an attempt to classify new data sets. Unsupervised learning </p><p>encompasses the detection of patterns within the data sets consisting of unlabeled data, i.e. </p><p>data sets with unspecified outputs, which fit to a general rule and can, therefore, be grouped </p><p>together. From an SHM viewpoint, unsupervised learning can be used, e.g., for detecting the </p><p>existence of damage through clustering of structural response data, while supervised learning </p><p>can advantageously be employed to detect the type and severity of damage [19]. Semi-</p><p>supervised learning, representing a combination of the two aforementioned learning schemes, </p><p>typically aims at obtaining a classification of data using both labeled and unlabeled data. </p><p>Semi-supervised learning schemes have been applied combined with other monitoring </p><p>techniques to extract information on modal characteristics of bridges [20]. </p><p>Since most SHM problems require inferring a function from labeled training data (e.g. to </p><p>assess the data or to predict new data), supervised learning is an appropriate means to solve </p><p>these problems. In supervised learning, the algorithms, according to [21], can be categorized </p><p>as logic-based algorithms (e.g., decision trees and rule-based classifiers), perceptron-based </p><p>algorithms or neural networks (e.g., single-layered perceptron, multi-layered perceptron and </p><p>radial basis function networks), statistical learning (e.g., naive Bayes classifiers and Bayesian </p><p>networks), instance-based learning (e.g., k-nearest neighbor algorithm), and support vector </p><p>machines. </p></li><li><p>4 </p><p>2.2 Prototype implementation of the machine learning approach </p><p>In this study, decentralized autonomous sensor fault detection is based on the principle of </p><p>analytical redundancy [22]: Instead of physically installing multiple sensors for measuring </p><p>one single parameter, analytical redundancy takes advantage of the redundant information </p><p>inherent in the SHM system and utilizes the coherences and relationships between the sensors </p><p>installed in the structure. It has been proven that the peak amplitudes of the frequency </p><p>spectrum, obtained by the Fourier transformation of acceleration response data, </p><p>corresponding to resonant response (i.e. modal peak amplitudes) from different sensors of the </p><p>same structure are correlated [23]. This correlation can be exploited to predict the modal peak </p><p>amplitudes of selected sensors, using the modal peak amplitudes of correlated sensors as </p><p>input data. Deviations between expected amplitudes and actual amplitudes (i.e. from the </p><p>measured data) are indicative of sensor faults and miscalibrations. Importantly, no a priori </p><p>knowledge about the structure or about the sensor instrumentation is required because, as a </p><p>purely data-driven approach, previously collected sensor data is taken as the sole basis for </p><p>fault detection. </p><p>A wireless SHM system is designed that comprises wireless sensor nodes, each of which </p><p>including an integrated 3-axis accelerometer, a base station, and a host computer. The </p><p>monitoring tasks executed by the SHM system are illustrated in Figure 1. During operation, </p><p>acceleration response data is sampled by each sensor node and locally transformed into the </p><p>frequency domain via an embedded Cooley-Tukey FFT algorithm. A peak detection </p><p>algorithm selects the highest peak of the frequency spectrum corresponding to the </p><p>fundamental eigenfrequency (modal peak amplitude), and the modal amplitudes are </p><p>communicated among the sensor nodes. Each sensor node predicts the modal amplitude of its </p><p>own acceleration response data (expected amplitude) using the modal peak amplitudes of </p><p>correlated sensor nodes and decides upon the existence of sensor faults based on deviations </p><p>between the expected and the actual modal peak amplitude. The outcomes of the fault </p><p>detection procedure of the sensor nodes are transmitted to the host computer via the base </p><p>station for storage and decision making. </p><p>Figure 1. Decentralized, autonomous fault detection procedure executed by the wireless SHM system </p></li><li><p>5 </p><p>The decentralized autonomous fault detection procedure proposed in this study relies on </p><p>the relationships among the modal peak amplitudes from different sensors. To map these </p><p>relationships an embedded machine learning approach with a supervised learning scheme is </p><p>introduced. To this end, artificial neural networks (ANNs) are designed and distributedly </p><p>embedded into each sensor node. As shown in Figure 2, the ANNs consist of three layers of </p><p>neurons: 1) an input layer of k neurons, 2) a hidden layer of m neurons to account for the non-</p><p>linear relationship among the modal peak amplitudes of different sensors [24], and 3) an </p><p>output layer of one neuron, which represents the predicted modal peak amplitude of the </p><p>sensor under consideration. The data is propagated through the ANN via the synapses (i.e. </p><p>connections between neurons), based on the weight of each connection. During the ANN </p><p>training, the weights of the synapses are adjusted until a selected set of input data results in </p><p>the desired output data. The ANN properties (i.e....</p></li></ul>

Recommended

View more >